Tailoring Datasets for Regioselectivity Predictions on Complex Substrates

The development of machine learning models to predict the regioselectivity of C(sp3)–H functionalization reactions is reported. A dataset for dioxirane oxidations was curated from the literature and used to generate a model to predict the regioselectivity of C–H oxidation. The performance of the model for a specific substrate could then be improved by adding additional datapoints that were selected using acquisition functions. A series of different acquisition functions were compared, and acquisition functions based on active learning were found to outperform those based on molecular and site similarity. Furthermore, it was found that smaller machine-designed datasets can give accurate predictions when larger, randomly-designed datasets fail. The use of acquisition functions for dataset enlargement significantly reduced the number of datapoints needed to perform accurate prediction. Finally, the workflow was shown to be applicable to predicting the regioselectivity of arene C–H radical borylation. These studies provide a quantitative alternative to the intuitive extrapolation from “model substrates” that is frequently used to estimate reactivity on complex molecules.